Unverified50% confidenceFactExact time
Gradient weighting on high-similarity negatives persistently harms downstream generalization, i.e., excessively pushing the model away from certain samples disrupts the overall structure of the embedding space, so retrieval performance drops rather than rises (termed 'gradient distortion').
1
Sources
50%
Confidence
Long-term
Relevance
10/14/2026
First Seen
Sources
Related Entities
Related Claims
VerifiedGradient descent iteratively adjusts weights in the opposite direction of the loss function's gradient, causing the prediction error to converge progressively.75% similarUnverifiedFine-tuning on data with narrow defects can sometimes cause models to become broadly misaligned.75% similarUnverified过度增强（如对俯仰角施加±90°随机扰动）会人为制造源域与目标域偏移，让模型见到推理时永远不会出现的极端视角，无益于泛化75% similarUnverifiedRepeatedly applying the same set of weights can easily cause gradients to explode or vanish across multiple iterations, similar to the gradient problems of early RNNs.74% similarUnverifiedCovariate shift and model drift can trigger negative transfer, causing additional data to drag down model performance.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1028174API
curl https://kongchang.com/api/v1/knowledge/claims/1028174MCP
get_claim(id=1028174)